A Fast Algorithm for Half-box Image Filtering

Abstract

Image averaging with a convolution kernel consisting of brick-wall cutoff along salient edges is useful for robust estimates of an image manifold and its color statistics, but their fast computation is non-trivial. Although some simplification of such averaging by using half-box kernels has been proposed, the conventional methods require multiple filtering processes. In this research, we propose a fast algorithm for image averaging using half-box convolution kernels. Our algorithm is based on summing the total colors within adjacent rectangular regions, where these total colors are efficiently obtained by applying a half-radius box filter only once via the moving average method. Our numerical experiments show that the proposed method is faster than conventional algorithms.